Prenatal Ambient Air Pollutant Mixture Exposure and Early School-age Lung Function : Environmental Epidemiology

Secondary Logo

Journal Logo

Original Research Article

Prenatal Ambient Air Pollutant Mixture Exposure and Early School-age Lung Function

Hsu, Hsiao-Hsien Leona,b; Wilson, Anderc; Schwartz, Joeld; Kloog, Itaia,b; Wright, Robert O.a,b; Coull, Brent A.e; Wright, Rosalind J.a,b,*

Author Information
Environmental Epidemiology 7(2):p e249, April 2023. | DOI: 10.1097/EE9.0000000000000249


What this study adds

Research linking prenatal air pollution with childhood lung function has largely considered one pollutant at a time. Real-life exposure is to mixtures of pollutants and their chemical components, and it is important to assess joint effects/effect modification by co-exposures. Our study leverages exposure from satellite data, predicting seven major air pollutants in a longitudinal pregnancy cohort to examine associations between prenatal exposure to air-pollution mixtures and children’s lung function. We applied a novel advanced data-driven statistical approach to explore sex-specific time-weighted associations between air pollution and lung function. Our findings suggest that increased prenatal O3, OC, and NH4+ were major contributors to reduced childhood lung function outcomes.


Pulmonary morphogenesis begins 3-4 weeks post-conception and continues after birth. Vulnerability to environmental pollutants begins in utero.1,2 The consequences of environmental damage on later life respiratory health are largely associated with loss of lung function. Lung function tracks an individual percentile established early in life, largely determining the level of maximal function reached as a young adult, which in turn affects the cumulative risk of lung disease as we age.3 Lung function at birth4 and lung function growth patterns established by approximately age 7 years determine early adult pulmonary function.5,6 Impaired adult maximal attained lung function is a major risk factor for chronic obstructive pulmonary disease (COPD), which is among the three leading causes of death globally and is a major driver of rising healthcare expenditures.7 An important step toward identifying those at risk for chronic lung disease is characterizing exposures and mechanisms that lead to and maintain early predisposition.

Starting in utero, ambient air pollution exposure, especially fine particulate matter (PM with an aerodynamic diameter ≤2.5 µm; PM2.5) has been linked with childhood respiratory outcomes,8–10 although results are inconsistent. Recent findings demonstrate a role for nitrate (NO3),11 a major component of PM2.5, and ground-level ozone (O3).12 However, studies have largely considered one pollutant or component at a time13 with a few studies including two pollutants in a single model to see which contributes independently to effects.9,14,15 Real-life exposure is to a mixture of pollutants and their chemical components.

PM2.5 is a mixture consisting of liquid droplets and solid particles with widely variable chemical composition sourced from natural phenomena (e.g., volcanoes) or anthropogenic activities (e.g., fossil fuel combustion)16,17 constituted from direct emission (primary) or secondary formation via chemical reactions such as nitrate (NO3), sulfate (SO42−),18 ammonium (NH4+),19 and organic compounds, and combustion process emissions from biomass burning, vehicles, soil, and road dust.20 Toxic effects of PM2.5 may be better characterized by considering its multiple chemical constituents.21 Failing to consider joint effects or effect modification by co-exposures can contribute to misleading results.22 Moreover, air pollutant mixtures may be dominated by one pollutant or another with variation across geographic areas depending on whether the mixture is generated primarily by local sources (e.g., local traffic or industries) or regional pollution containing transported primary and secondary pollutants. Heterogeneity in results may in part be explained by geographic variation in the chemical composition.23 Studies that consider multiple pollutants and their components as mixtures will further elucidate the impact of prenatal ambient pollution on early childhood lung function.

Moreover, analyses modeled at lower temporal resolution can result in missed or biased associations.24 Studies have shifted from considering exposure averaged over defined periods (e.g., entire gestation and trimesters) towards models utilizing data-driven methods coupled with highly temporally resolved exposure data (daily or weekly) to flexibly identify windows of effect.25 These studies highlight the advantages of more precisely identifying sensitive exposure windows to better elucidate underlying mechanisms.8,10,11,26–28 Exposure to air pollution over gestation can disrupt fetal lung growth and development with an impact across childhood and the lifespan.29 Sex-specific effects have also been demonstrated.8,10,11,27 Although methods that utilize highly temporally resolved exposure data advance our understanding, they allow consideration of one pollutant at a time. Other studies focusing on critical windows of exposure consider the individual constituents of PM2.5,30,31 but to our knowledge, none have examined multiple components together as a mixture along with other pollutants, particularly in conjunction with exposure modeling accounting for finer spatial and temporal resolution.32

Focusing on pollutants for which prior evidence exists for effects on lung function, we leverage available daily residence-based prenatal pollutant exposures estimated using satellite-based hybrid chemical-transport models, including nitrogen dioxide (NO2), ozone (O3), and fine particle constituents (elemental carbon, organic carbon [OC], nitrate [NO3], sulfate [SO42–], and ammonium [NH4+]) to examine associations between prenatal exposure to an air pollutant mixture and lung function in children enrolled in a prospective pregnancy cohort in the Northeastern United States. We implemented a novel statistical approach, Bayesian Kernel Machine Regression Distributed Lag Models (BKMR-DLMs),33 to flexibly estimate windows of vulnerability between prenatal exposure to an air pollution mixture and children’s pulmonary function and to explore interactions among mixture components as well as sex-specific effects.


Study participants

Subjects were from the Asthma Coalition on Community, Environment and Social Stress (ACCESS) project, an ethnically diverse urban pregnancy cohort designed to examine the effects of environmental exposures, starting prenatally, on childhood respiratory outcomes.34 Eligibility criteria included English- or Spanish-speaking women aged ≥18 years receiving routine prenatal care at recruitment sites. Eligible participants for this study (n = 483) were recruited from two Boston hospitals and affiliated community health centers at 24.1 ± 10.0 weeks gestation from August 2001 through December 2008. Of these, 220 children completed a home visit at age 6.99 ± 0.89 years during which pulmonary function was obtained; 205 (93%) provided acceptable spirometry among whom 198 also had air pollution exposure data. Participants included in these analyses did not differ from those eligible based on key covariates including child sex, maternal age at enrollment, maternal education, and maternal race/ethnicity (Supplemental Table S1; Procedures were approved by human studies committees at the Brigham and Women’s Hospital and Boston Medical Center and written consent was obtained from all mothers in their primary language of choice; assent was also obtained for children aged ≥ 7 years.

Daily prenatal air pollution mixture levels and temperature

The maternal residential address was geocoded and updated throughout gestation when participants moved as previously described.35 Daily residence-specific exposure to constituents of ambient fine particle exposure, including elemental carbon, OC, NO3, SO42–, and NH4+, was estimated using validated hybrid models incorporating chemical transport (GEOS-Chem) models and land-use regression models, calibrated with ground monitoring data to predict daily residence level exposure at 1 km × 1 km grid cells.32 Gaseous pollutants, including NO2 and O3, were estimated via a hybrid model using satellite data, a chemical transport model (GEOS-Chem), and land use variables, calibrated with ground monitoring data to make residence level predictions at 1 km × 1 km grid cells.36,37 A neural network was used for both to model nonlinearity and interactions between variables.

To account for seasonality and associated air pollution trends, daily temperature estimates were derived across gestation using a model that calibrated MODIS satellite surface temperature measurements to air temperature monitors using a hybrid land-use regression-based model.38 To reduce noise caused by day-to-day variation, weekly averages of all air pollution components and temperature were used in these analyses and were adjusted in all models. Greater detail on the methods and performance of the exposure models are provided in the online supplement;

Pulmonary function testing

Supplemental funding was obtained to assess spirometry in children conducted by trained research assistants using methods previously detailed.11,39 Before testing, child weight and height were measured to the nearest 0.1 kg, and 0.1 cm, respectively. Spirometry following American Thoracic Society guidelines40,41 with techniques modified for children ≤ 8 years of age42,43 was performed in participant homes with a portable MedGraphics laptop-based spirometer. Bronchodilators were held before testing. All tests were overseen for acceptability and reproducibility by a pediatric pulmonologist. Forced vital capacity (FVC, milliliters), forced expiratory volume in one second (FEV1, milliliters), and forced expiratory flow between 25% and 75% of the FVC (FEF25-75, milliliters per second) were recorded from a minimum of 3 (no more than 8) maneuvers. Lung function measures, height, and weight were approximately normally distributed. Spirometry parameters were regressed on age, sex, height, and race/ethnicity, then calculated residual values were divided by the standard deviation (SD) of the residuals to convert to z-scores with mean 0 and SD of 1. Notably, calculated z-scores including race/ethnicity were highly correlated with z-scores calculated not including race/ethnicity; Pearson correlations between the two z-scores were high for all spirometry parameters (FEV1 [0.97], FVC [0.98], FEV1/FVC [0.98], and FEF25–75 [0.98]).

Other covariates

Maternal race/ethnicity, age, and education were obtained via a questionnaire at enrollment. Race/ethnicity was categorized as Black (Black and Hispanic-Black), Hispanic (non-Black Hispanic), White (non-Hispanic White), and other/mixed. Mother’s perinatal smoking status was reported at baseline and/or in the third trimester of pregnancy as well as over the first postnatal year via interviews performed approximately every 3 months.

Statistical analysis

Our primary objectives were: (1) to identify critical windows during which the exposure to each component is associated with children’s lung function and (2) to estimate the exposure-response relationship while allowing for a potentially nonlinear and nonadditive relationship among the multiple components in the mixture and lung function.

We employed a newly developed model, BKMR-DLM, detailed elsewhere.33 BKMR-DLM is a form of kernel machine regression that takes as input repeated measures of each mixture component and uses weight functions that up- or down-weight exposure time periods, separately for each mixture component, during which that component has increased or decreased association with the outcome. We identify critical windows for mixture components as time periods that are up-weighted, whereas time periods with weight function near zero indicate little or no evidence of association between a component and the outcome at that time point. The BKMR-DLM then estimates a potentially nonlinear and nonadditive association between a health outcome and the constructed time-weighted exposures by using these weighted exposures as inputs in a BKMR model. Implementation of the method estimates the time-specific weights for each mixture component and the form of the kernel-based exposure-response function simultaneously.

We used the BKMR-DLM model applied to all exposures together to calculate a time-weighted, linear combination of the weekly exposures for each pollutant. We then used the time-weighted combination for each pollutant as a single component in subsequent mixture analyses. Because the vast majority of associations estimated from the BKMR-DLM were linear (with the exception of OC exposure and FEF25-75), to enhance interpretability we used the temporally-weighted exposures estimated from BKMR-DLM as inputs in multivariable linear models conducted in secondary analyses to provide point estimates and 95% confidence intervals of these linear slopes representing the association between each air pollution component and a lung function outcome. All pollutants weighted by BKMR-DLM are included in the same multivariate linear model. Models were fit for FEV1, FVC, FEF25-75, and the FEV1/FVC ratio. In addition to covariates accounted for in spirometry z-score calculations (age, sex, height, and race/ethnicity), we also adjusted for maternal age, education, and smoking status in both the BKMR-DLM and the multivariable linear models. The weekly averaged temperature is included in the BKMR-DLM. Therefore, it is temporally weighted along with the other pollutants to allow for the possibility of a temporally varying effect of temperature during pregnancy on the outcomes. Accordingly, these derived time-weighted temperature exposures are included as a covariate in the final multivariable linear models. We also fitted multivariable-adjusted models using a pregnancy-averaged air pollution mixture including the same covariates for comparison purpose. Analyses were conducted in R (v4.0.2, Vienna, Austria). A more detailed description of the modeling process is included in eMaterial;


Participant characteristics and exposure distributions

Table 1 summarizes participant characteristics as well as the distribution of spirometry outcomes and pollutants. The majority of mothers were Hispanic (63%) or Black (21%) and reported ≤12 years of education (67%). Figure 1 depicts prenatal exposure levels averaged over pregnancy for each component of the pollution mixture, demonstrating reasonable exposure variation. Figure 2 depicts Spearman’s correlation coefficients of the 40 weekly averaged exposure levels among the seven air pollutants. Weekly averaged mixture components were low to moderately correlated. Correlation matrices for the trimester-specific average exposures are also shown in Figure S1,

Table 1. - ACCESS participant characteristics (n = 198)
Child sex (n, %)
 Girls 93 47.0
 Boys 105 53.0
Race/ethnicity (n, %)
 Black/Hispanic Black 42 21.2
 Non-Black Hispanic 124 62.6
 Non-Hispanic White 18 9.1
 Other/mixed 14 7.1
Maternal education (n, %)
 >12 years 66 33.3
 ≤12 years 132 66.7
Child age at spirometry (years; mean, SD) 6.9 0.8
Child height at spirometry (cm; mean, SD) 121.6 7.2
Maternal age at enrollment (years; mean, SD) 27.2 5.9
Maternal smoking status (n, %)
 No smoking exposure 183 92.4
 Smoked prenatally, but not postnatally 6 3.0
 Did not smoke prenatally, but smoked postnatally 1 0.5
 Smoked both pre- and postnatally 8 4.0
Spirometry outcomes
 FEV1 raw value (L; mean, SD) 1.44 0.25
 FVC raw value (L; mean, SD) 1.58 0.30
 FEV1/FVC ratio (mean, SD) 0.92 0.06
 FEF25-75 raw value (L/s; mean, SD) 1.88 0.47
 z-score of FEV1 (mean, SD) a 0.00 0.99
 z-score of FVC (mean, SD) a −0.01 0.98
 z-score of FEV1/FVC ratio (mean, SD) a 0.02 1.00
 z-score of FEF25-75 (mean, SD) a 0.02 1.00
Prenatal air pollutant levels (mean, SD) b
 NO2 29.25 5.55
 O3 35.19 2.84
 EC 0.67 0.14
 OC 3.36 0.58
 NO3 1.34 0.27
 SO4 2− 3.40 0.35
 NH4 + 1.08 0.16
Prenatal temperature (Celsius; median, IQR) b 10.91 2.37
aAdjusted for age, sex, height, race/ethnicity.
bAveraged over entire pregnancy; unit for air pollutants is µg/m3, except NO2 and O3 is in ppb.

Figure 1.:
Averaged prenatal exposure levels for each component of the air pollution mixture, including EC, NH4 +, NO3 , NO2, O3, OC, and SO4 2−, in the study area estimated at participants’ residential location.
Figure 2.:
Spearman’s correlation coefficients between weekly average levels of air pollution mixtures. Boxplots represent the distributions of the correlations between each pair of two pollutants across the 40 gestational weeks.

Prenatal air pollution mixtures and child lung function

Figure 3 shows the week-specific weighting of each pollutant estimated by BKMR-DLMs, for FEV1 (panel A), FVC (panel B), FEV1/FVC ratio (panel C), and FEF25-75 (panel D). To examine the shape of the exposure-response relationship, we inspected the BKMR-DLM estimated associations and also applied BKMR to the estimated time-weighted exposures as a check.44 Results showed linear associations across the air pollution mixture components and spirometry outcomes (Figures S2 and S3,

Figure 3.:
Time-varying exposure weight functions estimated by BKMR-DLM with a quadratic kernel weighting prenatal air pollution mixture exposures and children’s spirometry outcomes (A) FEV1, (B) FVC, (C)FEV1/FVC, (D)FEF25-75 (top to down in order). The weight function is constrained and does not reflect the magnitude of the association or the direction of the association. It only reflects the timing of the association. All spirometry outcomes are z-scores adjusted for sex, age, height, race/ethnicity; models additionally adjusted for maternal age at enrollment, education level, weighted temperature, and prenatal smoking status. BKMR-DLM, Bayesian Kernel Machine Regression-Distributed Lag Models; FEF, forced mid-expiratory flow; FEV, forced expiratory volume; FVC, forced vital capacity.

Figure 4 shows the estimated linear slopes and 95% confidence intervals for each time-weighted exposure estimated from the BKMR-DLM and respective outcomes from multivariable linear regression models, considering the sample overall and stratified by child sex.

Figure 4.:
Multivariable-adjusted linear regression models predicting children’s spirometry outcomes (A) FEV1, (B) FVC, (C) FEV1/FVC, and (D) FEF25-75, using time-weighted prenatal mixture air pollution levels estimated by a BKMR-DLM. From left to right, effect estimates of each component in air pollution mixture including EC, NH4 +, NO3 , NO2, O3, OC, and SO4 2−, are shown. Effect estimates and 95% confidence intervals are plotted in the order of overall sample (all), boys, and girls to elucidate sex-specific associations. BKMR-DLM, Bayesian Kernel Machine Regression-Distributed Lag Models; FEF, forced mid-expiratory flow; FEV, forced expiratory volume; FVC, forced vital capacity.

For FEV1, the BKMR-DLM (Figure 3A) estimates identified early-to-mid pregnancy and late pregnancy as time periods of increased association between NH4+ and the outcome as evidenced by the larger weight function estimates in these time periods. Similarly, these models found that O3 had higher weighting at mid-pregnancy. In the multivariable-adjusted linear regression model that used the BKMR-DLM weighted exposure levels (Figure 4A), increased NH4+ was associated with decreased FEV1 in the overall sample (β = −0.82; P = 0.003; per SD increase in NH4+) and among boys (β = −0.97; P = 0.02); the association was in a similar direction in girls although this did not reach statistical significance (β = −0.65; P = 0.14). These findings suggested that O3 had similar effect on FEV1 in both boys and girls, but with a narrower confidence interval for boys.

The BKMR-DLM considering the FVC outcome (Figure 3B) also showed larger weights around early-to-mid pregnancy and late pregnancy for NH4+, and mid-pregnancy for O3. In the multivariable-adjusted linear regression model that used the BKMR-DLM weighted exposure levels (Figure 4B), increased NH4+ was associated with decreased FVC in the overall sample (β = −0.65; P = 0.03; per SD increase in NH4+) and this association was suggestive in boys (β = −0.82; P = 0.06), but not significant in girls (β = −0.36; P = 0.43).

For the FEV1/FVC ratio, the BKMR-DLM (Figure 3C) estimated higher weight for exposures around mid-pregnancy for O3, suggesting a significant sensitive window at 18-22 weeks gestation. In the multivariable-adjusted linear regression model that used the BKMR-DLM weighted exposure levels (Figure 4C), increased O3 was found associated with decreased FEV1/FVC ratio in the overall sample (β = −0.47; P = 0.005; per SD increase in O3) and in girls (β = −0.55; P = 0.01); while a similar pattern was found in boys, this was not statistically significant (β = −0.37; P = 0.19). There was also a suggested association between prenatal NO2 and a reduced FEV1/FVC ratio in the overall sample (P = 0.05) and between OC and a lower FEV1/FVC ratio in girls (P = 0.18) which did not reach statistical significance.

For the FEF25-75, the BKMR-DLM (Figure 3D) suggested a constant weighting across pregnancy for OC. In the multivariable-adjusted linear regression model that used the BKMR-DLM weighted exposure levels (Figure 4D), increased OC was associated with decreased FEF25-75 in the overall sample (β = −0.37; P = 0.01; per SD increase in OC) and in girls (β = −0.46; P = 0.03) with a similar pattern in boys that was not significant (β = −0.28; P = 0.24).

Comparison with the non-time-weighted model

To elucidate the advantage of performing BKMR-DLMs to derive time-weighted exposures that are associated with each outcome, we also compared the results of a multivariable-adjusted model using a pregnancy-averaged air pollution mixture (Figure S4, Using the pregnancy average exposures, the only significant association identified was between O3 exposure and a reduced FEV1/FVC ratio. Furthermore, in traditional multivariate regression analyses using trimester averages as exposure predictors (data not shown), we did not find any statistically significant associations between the prenatal air pollution mixture and child lung function parameters. Taken together, these findings indicate that weighted exposure derived from the BKMR-DLM improved the models’ ability to detect associations between air pollution mixtures and child lung function.

Interactions among air pollution mixture components

We investigated potential interactions using the BKMR-DLM model. Specifically, we looked at pairwise interactions among all pairs of exposures. To do this, we estimated the exposure-response function for a single component with all other components fixed at their median value. We then repeated this with one of the other components fixed at different quantiles (25%, 50%, and 75%). If the exposure-response functions of the first component are parallel across the quantiles of the second component, then there is no evidence of interactions. Deviation from parallel exposure-response functions indicates interaction. We found no evidence of interactions between the mixture components (results not shown). Therefore, we did not consider interactions in the linear models.


To our knowledge, this is the first prospective study to examine associations between maternal exposure to a mixture of multiple constituents of PM2.5 exposure and other gaseous pollutants (O3) and lung function in early school-aged children while identifying sensitive windows of exposure. These analyses demonstrated that particular components of the mixture, specifically O3, NH4+, and OC, had the strongest influence on lung function outcomes. Prenatal exposure to elevated O3 levels during mid-pregnancy was associated with a reduced FEV1/FVC ratio consistent with an obstructive pattern, especially among girls. Increased OC was associated with decreased FEF25-75 indicative of peripheral airway disease, particularly among girls. Boys born to mothers with higher prenatal NH4+ exposure in the early-mid and late pregnancy periods were more likely to have reductions in both FEV1 and FVC, suggesting an impact on lung growth overall.

These analyses coupled weekly-averaged multi-pollutant exposure data across pregnancy combined with novel advanced data-driven statistical modeling to obtain relative weighting of the exposure of each mixture component at each time point throughout gestation to identify pollutant-specific windows of vulnerability, rather than using predefined or convenient time periods (e.g., pregnancy or trimester averaged exposure) that can result in missed or biased associations.24,33 These methods also allowed us to estimate potential sensitive windows that may coincide with specific embryologic stages of respiratory system development. Overlaying identified windows of exposure for an environmental mixture upon our current knowledge about the progression of developmental events in utero may provide a more definitive understanding of the effects of air pollutants on offspring outcomes. For example, this approach can inform future studies investigating the underlying cellular and physiologic mechanisms being perturbed in the identified windows of susceptibility.

There has been increased interest in ozone effects on lung function given the complex relationships between O3 and other pollutants with documented effects on respiratory health. Long-term O3 exposure in children has been associated with decreases in lung function, most consistently with reductions in FEV1.45 Only one prior study considered prenatal exposure to O3. Latzin et al.46 examined associations between O3 averaged over pregnancy and lung function assessed during unsedated sleep in newborns aged 5 weeks (tidal breathing, lung volumes, and exhaled nitric oxide) and did not find significant associations. Moreover, few studies have included co-pollutant control. The current multi-pollutant analyses leveraging highly temporally resolved exposure data found that mid-pregnancy O3 exposure was associated with a decreased FEV1/FVC ratio. These findings are consistent with studies in children suggesting that growth of FEV1 may be particularly susceptible to ozone effects whereas a relationship between FVC and ozone has not been clearly demonstrated.45

Associations noted in relationship to NH4+ exposure are interesting as this pollutant has received less attention than other secondarily formed aerosols (OC, SO42–). Although NH4+ is generally considered to be non-toxic, it has been found to be an airway irritant in adult studies and has been receiving increased attention in relation to other health effects.23 For example, a cross-sectional study in adults enrolled in the China Pulmonary Health study linked exposure to particle-bound NH4+ with reduced FEV1.31 Other studies focusing on prenatal exposure to NH4+ have linked exposure to anthropometric outcomes in children. For example, one study in China, using trimester averaged exposure levels, demonstrated that exposures to NH4+ in the 2nd and 3rd trimesters were associated with reduced growth trajectories and lower BMI z-scores in children followed to age 6 years.30 Interestingly, these identified windows are relatively consistent with what we found for the relationship between prenatal NH4+ exposure and reductions in both FEV1 and FVC in our data. It is possible that exposure to this pollutant impacts underlying processes involved in overall somatic growth with consequent effects on lung growth as well. Several carefully timed processes can be disrupted by prenatal environmental exposures influencing lung development, highlighting the importance of understanding the timing of exposure as well as the dose.29 Our findings of increased exposure weighting of NH4+ in early gestation identified by the BKMR-DLM coincides with the pseudo-glandular stage of lung development.47 This stage is critical for the formation of supporting structures of the airway, including smooth muscle, mucosal glands, and vasculature responsible for the exchange of oxygen and nutrients.47 It is possible that toxin-related disruption experienced during this embryologic time point may restrict lung growth by compromising the formation of supporting structures necessary to sustain adequate growth going forward. Further, studies have hypothesized that prenatal particle exposure during pregnancy may result in placental inflammation which can disrupt placental function leading to inadequate placental perfusion and impaired transplacental nutrient exchange, which also can alter fetal somatic growth more generally.48

Our modeling framework also enhances our ability to identify underlying effect modifications.25 Although we did not have a sufficient sample size to conduct sex-stratified BKMR-DLMs, our sex-stratified analyses in multivariable-adjusted linear regressions suggested that boys may be more vulnerable to prenatal exposure to NH4+ whereas girls may be more susceptible to O3 and OC exposures. Interestingly, these linear regression analyses showed that the effect of O3 on reduced FEV1/FVC was more evident among girls as compared to boys. Notably, a sex-specific effect has been reported in rodent models, where female off-spring of prenatally O3-exposed rats had increased pulmonary inflammation and higher nonallergic overall inflammatory markers compared with air controls,49 suggesting a potential pathway of immuno-dysregulation for female offspring.

This study has several strengths. It is important to examine multiple pollutant effects on lung health in lower-income, ethnically diverse urban samples such as this one, as they are more highly exposed to ambient air pollution as well as being more likely to have reduced lung function. Combining more highly temporally resolved exposure data with a novel BKMR-DLM enabled us to estimate data-driven exposure-response weighting for each air pollution mixture at each time point in the same model accounting for other co-exposures.33 We also acknowledge some limitations. We may have lacked the power to detect all associations, particularly interactions among components in the mixture. Our methodology may also raise the concerns of an increased possibility of false positive errors from multiple comparisons as a result of modeling multiple exposures at many timepoints, however, our method is performing dimension reduction by smoothing the exposure effects across the multiple exposure time points and essentially reduces the risk of false positive findings. Another limitation is that, as in any mixture analysis, methods proposed in this study considering multi-pollutant mixture effects and their associations may vary by the composition of considered mixture components. As always, there is potential for confounding by co-exposures that are not included in the model. Our results should be interpreted in light of this limitation. Also, one other consideration is that direct exposure to air pollution over childhood can also impact lung growth and development. These analyses focus on the impact of maternal exposure over gestation on fetal lung development. We plan to extend the air pollution models out over further years to derive exposure estimates that enable assessment of postnatal exposure to an air pollutant mixture over childhood, which is conceptualized as a distinct exposure because the child is directly breathing the air. These future analyses can also consider the cumulative effects of pre- and postnatal exposure but at this time, considering postnatal exposure to the air pollution mixture is beyond the scope of our analyses.

Further studies with larger sample sizes are warranted to more fully elucidate the sex-specific effects of prenatal air pollution mixtures on lung function outcomes and to have greater power to detect interactions. Also, our subjects are recruited from two hospitals in the northeastern US, results may not generalize to other sociodemographic groups or geographic areas as the makeup of air pollution and components driving effects may vary.

In conclusion, these analyses answer a recent call to expand studies considering the context of air pollution mixtures when examining respiratory health effects.13 Future studies in samples drawn from greater geographic coverage will allow researchers to examine how associations may vary based on the proportional contribution of pollutants or constituents to the mixture. Such studies can point to which pollutant (or pollutants) is most strongly associated with the outcome of interest in a given region, which can then better inform intervention strategies. These findings may also have broader implications for health across the lifespan, as early life lung function influences adult cardiorespiratory disease as well as longevity.4,50 In light of the accumulating evidence linking prenatal air pollution and longer-term respiratory morbidity,51 public policies aimed at reducing exposure among pregnant women are needed to promote optimal lung development and prevent respiratory illness in later life.


1. Dratva J, Zemp E, Dharmage SC, et al. Early life origins of lung ageing: early life exposures and lung function decline in adulthood in two European cohorts aged 28-73 years. PLoS One. 2016;11:e0145127.
2. Sly PD, Bush A. Environmental contributions to respiratory disease in children. Kendig’s Disorders of the Respiratory Tract in Children. 2019:49–56.e3.
3. Jordan BK, McEvoy CT. Trajectories of lung function in infants and children: setting a course for lifelong lung health. Pediatrics. 2020;146:e20200417.
4. Stern DA, Morgan WJ, Wright AL, Guerra S, Martinez FD. Poor airway function in early infancy and lung function by age 22 years: a non-selective longitudinal cohort study. Lancet. 2007;370:758–764.
5. Bush A. Too little, too late: adult lung disease cannot be prevented by interventions in adult life. Am J Respir Crit Care Med. 2023;207:124–126.
6. Morgan WJ, Stern DA, Sherrill DL, et al. Outcome of asthma and wheezing in the first 6 years of life: follow-up through adolescence. Am J Respir Crit Care Med. 2005;172:1253–1258.
7. Duan KI, Birger M, Au DH, Spece LJ, Feemster LC, Dieleman JL. Health care spending on respiratory diseases in the United States, 1996-2016. Am J Respir Crit Care Med. 2023;207:183–192.
8. Hsu HH, Chiu YH, Coull BA, et al. Prenatal particulate air pollution and asthma onset in urban children. identifying sensitive windows and sex differences. Am J Respir Crit Care Med. 2015;192:1052–1059.
9. Wright RJ, Hsu HL, Chiu YM, et al. Prenatal ambient ultrafine particle exposure and childhood asthma in the Northeastern United States. Am J Respir Crit Care Med. 2021;204:788–796.
10. Lee AG, Le Grand B, Hsu HL, et al. Prenatal fine particulate exposure associated with reduced childhood lung function and nasal epithelia GSTP1 hypermethylation: Sex-specific effects. Respir Res. 2018;19:76.
11. Bose S, Rosa MJ, Chiu YHM, et al. Prenatal nitrate air pollution exposure and reduced child lung function: timing and fetal sex effects. Environ Res. 2018;167:591–597.
12. Liu X, Fu L, Yang X, Wang Z. Exposure to O3 during pregnancy and offspring asthma induced by OVA: sensitive window identification. Environ Pollut. 2021;270:116297.
13. Garcia E, Rice MB, Gold DR. Air pollution and lung function in children. J Allergy Clin Immunol. 2021;148:1–14.
14. Gehring U, Gruzieva O, Agius RM, et al. Air pollution exposure and lung function in children: the ESCAPE project. Environ Health Perspect. 2013;121:1357–1364.
15. Lavigne E, Donelle J, Hatzopoulou M, et al. Spatiotemporal variations in ambient ultrafine particles and the incidence of childhood asthma. Am J Respir Crit Care Med. 2019;199:1487–1495.
16. Adams K, Greenbaum DS, Shaikh R, van Erp AM, Russell AG. Particulate matter components, sources, and health: systematic approaches to testing effects. J Air Waste Manag Assoc. 2015;65:544–558.
17. Kelly FJ, Fussell JC. Size, source and chemical composition as determinants of toxicity attributable to ambient particulate matter. Atmos Environ. 2012;60:504–526.
18. Behera SN, Sharma M. Reconstructing primary and secondary components of PM2.5 composition for an urban atmosphere. Aerosol Sci Technol. 2010;44:983–992.
19. Park R, Jacob D, Field B, Yantosca R. Natural and transboundary pollution influences on sulfate-nitrate-ammonium aerosols in the United States: implications for policy. J Geophys Res. 2004;109,D15204.
20. Oanh KNT, Thiansathit W, Bond TC, et al. Compositional characterization of PM2.5 emitted from in-use diesel vehicles. Atmos Environ. 2010;44:15–22.
21. Bell ML, Dominici F, Ebisu K, Zeger SL, Samet JM. Spatial and temporal variation in PM(2.5) chemical composition in the United States for health effects studies. Environ Health Perspect. 2007;115:989–995.
22. Braun JM, Gennings C, Hauser R, Webster TF. What can epidemiological studies tell us about the impact of chemical mixtures on human health?. Environ Health Perspect. 2016;124:A6–A9.
23. Masselot P, Sera F, Schneider R, et al. Differential mortality risks associated with PM2.5 components: a multi-country, multi-city study. Epidemiology. 2022;33:167–175.
24. Wilson A, Chiu YM, Hsu HHL, Wright RO, Wright RJ, Coull BA. Potential for bias when estimating critical windows for air pollution in children’s health. Am J Epidemiol. 2017;186:1281–1289.
25. Wilson A, Chiu YM, Hsu HHL, Wright RO, Wright RJ, Coull BA. Bayesian distributed lag interaction models to identify perinatal windows of vulnerability in children’s health. Biostatistics. 2017;18:537–552.
26. Chang HH, Reich BJ, Miranda ML. Time-to-event analysis of fine particle air pollution and preterm birth: results from North Carolina, 2001-2005. Am J Epidemiol. 2012;175:91–98.
27. Chiu YHM, Hsu HHL, Coull B, et al. Prenatal particulate air pollution and neurodevelopment in urban children: examining sensitive windows and sex-specific associations. Environ Int. 2016;87:56–65.
28. Warren J, Fuentes M, Herring A, Langlois P. Spatial-temporal modeling of the association between air pollution exposure and preterm birth: identifying critical windows of exposure. Biometrics. 2012;68:1157–1167.
29. Veras MM, Oliveira Alves N, Fajersztajn L, Saldiva PHN. Before the first breath: prenatal exposures to air pollution and lung development. Cell Tissue Res. 2016;367:445–455.
30. Sun X, Liu C, Liang H, et al. Prenatal exposure to residential PM2.5 and its chemical constituents and weight in preschool children: a longitudinal study from Shanghai, China. Environ Int. 2021;154:106580.
31. Yang T, Chen R, Gu X, et al.; China Pulmonary Health Study Group. Association of fine particulate matter air pollution and its constituents with lung function: the China pulmonary health study. Environ Int. 2021;156:106707.
32. Di Q, Koutrakis P, Schwartz JD. A hybrid prediction model for PM2.5 mass and components using a chemical transport model and land use regression. Atmos Environ. 2016;131:390–399.
33. Wilson A, Hsu H-HL, Chiu Y-HM, et al. Kernel machine and distributed lag models for assessing windows of susceptibility to environmental mixtures in children’s health studies. Ann Appl Stat. 2022;16:1090–1110.
34. Wright RJ, Suglia SF, Levy J, et al. Transdisciplinary research strategies for understanding socially patterned disease: the Asthma Coalition on Community, Environment, and Social Stress (ACCESS) project as a case study. Cien Saude Colet. 2008;13:1729–1742.
35. Brunst KJ, Sanchez-Guerra M, Chiu YM, et al. Prenatal particulate matter exposure and mitochondrial dysfunction at the maternal-fetal interface: effect modification by maternal lifetime trauma and child sex. Environ Int. 2018;112:49–58.
36. Di Q, Amini H, Shi L, et al. Assessing NO2 concentration and model uncertainty with high spatiotemporal resolution across the contiguous United States using ensemble model averaging. Environmental science & technology. 2019;54:1372–1384.
37. Di Q, Rowland ST, Koutrakis P, Schwartz JD. A hybrid model for spatially and temporally resolved ozone exposures in the continental United States. J Air Waste Manag Assoc. 2017;67:39–52.
38. Kloog I, Nordio F, Coull B, Schwartz JD. Predicting spatiotemporal mean air temperature using MODIS satellite surface temperature measurements across the Northeastern USA. Remote Sens Environ. 2014;150:132–139.
39. Lee AG, Chiu YM, Rosa MJ, et al. Association of prenatal and early childhood stress with reduced lung function in 7-year-olds. Ann Allergy Asthma Immunol. 2017;119:153–159.
40. Beydon N, Davis SD, Lombardi E, et al.; American Thoracic Society/European Respiratory Society Working Group on Infant and Young Children Pulmonary Function Testing. An official American thoracic society/European respiratory society statement: pulmonary function testing in preschool children. Am J Respir Crit Care Med. 2007;175:1304–1345.
41. Miller MR, Hankinson J, Brusasco V, et al. Standardisation of spirometry. Eur Respir J. 2005;26:319.
42. Arets HGM, Brackel HJL, van der Ent CK. Forced expiratory manoeuvres in children: do they meet ATS and ERS criteria for spirometry?. Eur Respir J. 2001;18:655–660.
43. Eigen H, Bieler H, Grant D, et al. Spirometric pulmonary function in healthy preschool children. Am J Respir Crit Care Med. 2001;163(3 Pt 1):619–623.
44. Bobb JF, Valeri L, Claus Henn B, et al. Bayesian kernel machine regression for estimating the health effects of multi-pollutant mixtures. Biostatistics. 2015;16:493–508.
45. Holm SM, Balmes JR. Systematic review of ozone effects on human lung function, 2013 through 2020. Chest. 2022;161:190–201.
46. Latzin P, Röösli M, Huss A, Kuehni CE, Frey U. Air pollution during pregnancy and lung function in newborns: a birth cohort study. Eur Respir J. 2009;33:594–603.
47. Tang W, Du L, Sun W, et al. Maternal exposure to fine particulate air pollution induces epithelial-to-mesenchymal transition resulting in postnatal pulmonary dysfunction mediated by transforming growth factor0²/Smad3 signaling. Toxicol Lett. 2017;267:11–20.
48. Kannan S, Misra DP, Dvonch JT, Krishnakumar A. Exposures to airborne particulate matter and adverse perinatal outcomes: a biologically plausible mechanistic framework for exploring potential effect modification by nutrition. Environ Health Perspect. 2006;114:1636–1642.
49. Sharkhuu T, Doerfler DL, Copeland CB, Luebke RW, Gilmour MI. Effect of maternal exposure to ozone on reproductive outcome and immune, inflammatory, and allergic responses in the offspring. J Immunotoxicol. 2011;8:183–194.
50. Carraro S, Scheltema NM, Bont L, Baraldi E. Early-life origins of chronic respiratory diseases: understanding and promoting healthy ageing. Eur Respir J. 2014;44:1682–1696.
51. Vieira SE. The health burden of pollution: the impact of prenatal exposure to air pollutants. Int J Chron Obstruct Pulmon Dis. 2015;10:1111–1121.

Prenatal exposure; Multi-pollutant; Mixture; Lung function; spirometry

Supplemental Digital Content

Copyright © 2023 The Authors. Published by Wolters Kluwer Health, Inc. on behalf of The Environmental Epidemiology. All rights reserved.